AI-generated product recommendations: new advertising opportunities?

AI-generated product recommendations are personalised product suggestions assembled by machine learning models based on user behaviour, search intent and purchase history. In the context of Google Ads and Meta Ads, they give advertisers a direct opportunity to show the right product to the right person at exactly the right moment, without a campaign manager having to make every selection manually.

Key takeaways

• AI-generated product recommendations combine user signals, inventory data and ad formats into personalised campaigns that scale automatically.
• Google Shopping, Performance Max and Meta Dynamic Product Ads are the primary channels where AI recommendations drive direct revenue.
• A high-quality product feed is the foundation: without accurate feed data, no AI recommendation system performs optimally.
• AdBrains uses proprietary AI technology to synchronise product feeds, bidding strategies and audience signals daily for maximum ROAS.
• Both e-commerce stores and lead generation businesses can benefit, provided the recommendation logic is aligned with the right conversion point.

What exactly are AI-generated product recommendations?

AI-generated product recommendations are suggestions determined by algorithms using multiple data points simultaneously: which products a user previously viewed, what similar users purchased, what stock is available, and which product carries the highest margin. In advertising systems like Google Ads, these recommendations are translated into dynamic ad content, where the system automatically decides which product from a feed is most relevant for a specific user at a specific moment.

The difference from traditional product advertising is fundamental. In a classic Shopping campaign, the advertiser decides in advance which products to promote, what budget to allocate and which audience to target. AI recommendation systems invert this: the system continuously analyses all available signals and autonomously decides which product has the highest conversion probability for which user. This makes the approach both more scalable and, in practice, more effective for large product catalogues.

Which advertising channels benefit most?

Not every advertising channel is equally suited to AI-generated product recommendations. The greatest impact is visible on channels that are closely integrated with product feeds and support dynamic ad content.

Google Shopping and Performance Max (PMax) are the strongest positioned. Performance Max combines all Google channels, including Search, Display, YouTube and Gmail, into one campaign. The system uses the product feed as its foundation and automatically optimises which product is shown on which channel to which user. According to Google Ads Help (2026), Performance Max uses machine learning to test asset combinations and audiences, automatically showing the best-performing combinations more frequently.

Meta Dynamic Product Ads (DPA) work on a similar principle. You connect a product catalogue to your campaign, and Meta decides based on user behaviour within the platform which product is relevant for which user. This is particularly powerful for remarketing: a visitor who viewed a product on Elletens.nl without purchasing sees that specific product again in their Facebook or Instagram feed.

The product feed as the engine of the system

An AI recommendation system is only as strong as the data it receives. The product feed is the absolute core. An incomplete or outdated feed leads to irrelevant recommendations, missed opportunities and wasted budget. For webshops like E-4motion.com, which sells new electric folding bikes, this means every product in the feed must contain complete and current information: correct categories, accurate titles with relevant search terms, current prices, available variants and high-quality product images.

The key elements of a strong product feed for AI recommendations are:

• Product title: contains the primary keyword and relevant specifications (brand, model, colour, size).
• Product category (Google Product Category): correctly categorised so the algorithm understands the right context.
• Price and availability: synchronised in real time with the webshop, so out-of-stock products are never advertised.
• GTIN / EAN codes: helps Google connect the product to existing search demand and competitive data.
• Custom labels: for margin segmentation, seasonal relevance or bestsellers, allowing the AI system to set priorities.
• Product description: rich in relevant keywords without keyword stuffing, so the system interprets the content correctly.

In our practice, we see that optimising product feed quality for e-commerce clients like ToetsJeKennis.nl (online exams and courses) has a direct impact on the quality of AI selections. When feed titles, descriptions and categories do not match the actual search intent of the target audience, the AI recommendations become less relevant as a result.

How AdBrains AI automates and optimises product recommendations

AdBrains has developed its own AI technology that goes beyond the standard settings of Google or Meta. Where platform algorithms primarily optimise on short-term signals within their own ecosystem, the AdBrains AI combines external and internal data points to make structurally better product selections.

The system operates with several specialised modules working in concert. First, the server-side signal enrichment module ensures that conversion signals are not only sent via the standard Google tag, but are enriched with first-party data via a proprietary server-side Google Tag Manager (sGTM) infrastructure. This is crucial for AI recommendations: the more and better conversion signals the Smart Bidding algorithm receives, the more accurately the system learns which products generate the highest conversion value for which audiences.

Second, the automatic tCPA/tROAS optimisation module adjusts bidding strategies daily per product group. For a webshop like E-4motion.com, where the margin per model of electric folding bike differs, it is essential that higher-margin products justify higher bids. The AdBrains system connects margin data directly to the bidding strategy, so the AI optimises not just on volume but on actual profitability.

The Keyword Incubator also plays a role in product recommendations via Search. New products or categories are first tested in an isolated incubator campaign before being promoted to the production campaign. This prevents a new product from immediately competing for budget against proven bestsellers, and gives the system time to learn which search terms generate the most value for that specific product.

The multi-agent verification system forms the quality layer above all these modules. Four independent AI agents verify every optimisation decision before it is executed. This means that any automatic adjustment to a product group bid, a negative keyword or a feed optimisation is always verified for accuracy and impact before anything changes in campaign management. In practice, this results in a much lower risk of errors that frequently occur in fully automated systems without verification.

Finally, the audience management automation module manages audiences weekly based on product interaction. Visitors who viewed a specific product without purchasing are automatically added to an RLSA audience and receive dynamic remarketing featuring exactly that product, supplemented by relevant models based on the AI recommendation logic.

From theory to practice: a campaign setup for AI recommendations

A concrete campaign setup for AI-generated product recommendations involves a number of consistent steps. The table below provides an overview of the most effective approach per advertising channel.

When setting up such a campaign structure, it is important to start with a clear conversion definition. What counts as a valuable conversion: a purchase, a test ride request, a quote form submission? Without a unambiguous conversion point, no AI system can effectively learn which products or services deliver the best results.

Common mistakes with AI product recommendations

There are several pitfalls advertisers regularly encounter when deploying AI recommendations in their campaigns. Avoiding these mistakes makes the difference between a campaign that learns and scales, and one that consumes budget without clear direction.

• Incomplete or outdated product feed: the algorithm can only recommend what it sees in the feed. Missing attributes lead to less relevant recommendations.
• Audience signals that are too broad at the start: give the system concrete signals from the outset (existing converters, customer lists) rather than fully trusting the algorithm for prospecting.
• No margin segmentation: not every product deserves the same ad budget. Use custom labels to prioritise high-margin products.
• Optimising too early: AI recommendation systems need data to learn. Do not adjust bidding strategies before the system has collected sufficient conversion data (generally a minimum of 30-50 conversions per campaign per month).
• Lack of negative keywords in Search: even in an AI-driven system, irrelevant search terms can distort product recommendations. Daily search term mining is essential.

Frequently asked questions about AI-generated product recommendations

Do AI product recommendations work for small webshops with a limited range?

Yes, small webshops can also benefit, but the dynamics differ. AI recommendation systems work best when there is sufficient variety in the assortment and enough conversion data available. For a webshop with ten products, a fully dynamic system is less meaningful than for a catalogue of hundreds of products. Small webshops benefit more from a well-optimised static campaign structure combined with targeted remarketing than from a fully dynamic AI recommendation system.

What is the difference between Performance Max and a standard Shopping campaign for product recommendations?

Performance Max uses the product feed as its foundation but distributes recommendations across all Google channels simultaneously, including YouTube, Display, Gmail and Search. A standard Shopping campaign is limited to the Shopping network and Google Search. PMax gives the algorithm more freedom and more channels to test, which can lead to a broader reach. A standard Shopping campaign gives the advertiser more control over which products are promoted and via which search terms. For large catalogues with sufficient conversion data, PMax typically performs better. For niche assortments, a more manual approach may sometimes be preferable depending on the specific situation.

How do I ensure AI recommendations do not only promote the cheapest products?

The algorithm optimises by default on conversions or conversion value, not on margin. If cheap products generate more clicks and conversions, they are automatically recommended more often. To correct this, use custom labels in the product feed to indicate margin segments, and set Target ROAS (tROAS) at a level that accounts for the profit margin per product category. AdBrains connects margin data directly to the bidding strategy, so the system steers on actual profitability rather than just on revenue volume.

Is server-side tracking really necessary for AI product recommendations?

Server-side tracking is not strictly required, but has a major influence on the quality of AI recommendations. Browser-based conversion tracking misses a significant portion of conversions due to ad blockers, privacy measures and iOS restrictions. This means the Smart Bidding algorithm works with incomplete data, directly affecting recommendation quality. With server-side tracking via a proprietary sGTM setup, more conversions are measured, signals are enriched with first-party data, and the algorithm learns faster and more accurately. In our practice, we see that Enhanced Conversions combined with server-side tracking structurally increases measured conversions compared to a purely client-side setup.